1
On the Relationship between Phonology and Phonetics
Eugeniusz Cyran
Należy fonetykę od fonologii odróżniać, ale nie należy ich oddzielać.
Phonetics and phonology should be told apart, but not taken apart [EC].
(Stieber 1955: 73)
1. Introduction
In his brief note, Stieber lays out his views on the relationship between the two
components of language by saying that “phonological considerations which are
not based on phonetic studies hover in the air... on the other hand, phonetic studies
which do not aim at a phonological synthesis are practically pointless [EC]”
1
At
first, the above quotations look paradoxical. How can two things be told apart if
they cannot be taken apart, and studied separately? What are the criteria for
deciding that a given phenomenon is phonological or phonetic? What is the nature
of their relationship if phonology and phonetics are indeed autonomous?
Any discussion of the relationship between phonology and phonetics assumes
implicitly that these are indeed two separate entities. Stieber warns us, however,
against two opposite perspectives in the way phonological or phonetic studies can
be carried out which he considers bad practice. Phonologists are advised to
ground their work in phonetics, while phoneticians are cautioned that their
endeavours should serve higher purpose – phonological synthesis, whatever that
is. More than half a century later, the above questions are still relevant and the
debate concerning the relation between phonology and phonetics is even further
away from a solution. A variety of points of view exist, including those which
exclude one of the two components from grammar. Consequently, the very
question of the relationship between phonology and phonetics becomes
immaterial. Thus, on the one hand, there is a strong position of Ohala (1990), who
maintains that there is no interface between phonology and phonetics because
phonetic theory itself is sufficient to deal with the observed sound patterns in
languages. Though largely correct – phonetic theory is indeed making constant
progress and is outstripping phonology in more and more areas to do with the
organization and behaviour of speech sounds – Ohala still seems to distinguish the
1
“Rozważania fonologiczne nie oparte na badaniach fonetycznych wiszą w powietrzu... zaś
badania fonetyczne nie dążące do fonologicznej syntezy są właściwie bezcelowe” (Stieber 1955:
73). It was professor Piotr Ruszkiewicz who drew my attention to this quote.
2
two fields, if only terminologically, by saying, for example, that “...phonetics
offers one of the most obvious paths between phonology and other disciplines” (p.
165). What is phonology then? It appears that by refuting the existence of an
interface between phonology and phonetics, and advocating a close integration
between the two fields, Ohala is no less paradoxical than Stieber. Or, to put it
differently, he might be talking about the same thing as Stieber, though from a
strictly phonetic perspective which is covering more and more ground in the
‘universe of speech’, allowing for phonological synthesis, yet not offering a sharp
definition of what phonology is, or should be. Ohala, then, responds positively to
the postulate that phonology and phonetics should not be taken apart, and offers
only a phonetic perspective on how to tell them apart.
On the other extreme, we find proposals such as Hale and Reiss (2000, 2008)
who draw a sharp line between form and substance, excluding phonetic substance
from phonology, and arguing that the latter is a computational module of
grammar, while the former is not.
2
Their position is sharply an eloquently laid out
in the following quotes, of which the first one seems to be compatible with the
views of Stieber. The second one, however, suggests that phonology and
phonetics should be taken apart, or does it?
The modular approach to linguistics, and to science in general, requires that we
both model the interactions between related domains and sharply delineate one
domain from the other (Hale and Reiss 2000: 158).
Phonology is not and should not be grounded in phonetics since the facts that
phonetic grounding is meant to explain can be derived without reference to
phonology. Duplication of the principles of acoustics and acquisition inside the
grammar violates Occam’s razor and thus must be avoided. Only in this way will
we be able to correctly characterize the universal aspects of phonological
computation (p. 162).
On a closer inspection, what Hale and Reiss say is not incompatible with either
Ohala’s or Stieber’s views. It is simply a different, phonological, perspective. One
that does not ignore the results of phonetic research. On the contrary, it seems to
embrace it happily because the core of the substance-free research programme in
phonology is that substance-based speech sound patterns should have a phonetic
explanation only. Consequently, pure phonology is a computational module which
is much smaller than it is generally assumed, but it does exist as separate from
2
For a recent survey of a broader range of proposals see, e.g. (Kingston 2007).
3
phonetics. An additional and long-standing argument in favour of substance-free
phonology mentioned by Hale and Reiss is based on the fact that phonology must
be modality free as there is such a thing as the phonology of signed language. All
this, however, does not mean that some way of relating phonetics with phonology,
just as signs and phonology, should not be sought. Whether this type of
phonological practice ‘hovers in the air’ is then an empirical question.
In this paper, I attempt to fully embrace the spirit of the views of both Stieber
and Ohala by working from a phonological perspective similar to Hale and
Reiss’s. The seeming paradox in the first quote in this paper calls for a no less
paradoxical solution. I start with the assumption that phonology and phonetics
cannot be told apart (delineated) if they are not taken apart first. How they interact
is another issue, which will also be addressed. Firstly, I begin from a phonological
perspective of Government Phonology (GP), which seems to be a good candidate
for a substance-free model, if some modifications are implemented. Secondly, I
assume phonology and phonetics to be autonomous (told apart) yet interacting in a
conventionalized way to form a sound system (not taken apart). It will be argued
that most of the confusion in the discussion on phonology and phonetics stems
from the fact that sound systems are mistaken for phonology.
2. Sound system, phonology and phonetics
In the ‘universe of speech’, a sound system is the sum total of phonological and
phonetic aspects which together are responsible for the observed phonetic facts. In
this model, a sound system cannot be identified with phonology, because that
would ignore phonetics. Neither can a sound system be identified with phonetics
only. In other words, a sound system stands behind the observed phonetic facts in
a given language, but it cannot be directly identified with phonetics. As a
consequence, phonetically observed facts are not entirely independent of the
particular system in which they occur. Phonetic facts are always a result of
phonetic interpretation of phonological representation. They follow from the
system, and as such they may be ambiguous and misleading. To understand a
sound system, one has to find out how phonology and phonetics interact in that
system. Both phonology and phonetics are separate and can be studies separately,
but when sound patterns or systems are taken into account, the two aspects must
dove-tail to produce the results. The graph in (1) and the discussion below further
clarify how phonology relates to phonetics in a sound system.
4
(1)
Sound System
=
Phonology
+
Phonetics
(grammar-internal)
(grammar-external?)
Representation & Computation
Phonetic interpretation
- privative categories
- universal principles
- (un)licensing, spreading
- language / system specific
- (de)composition
conventions (rules)
- sociolinguistic modifications
2.1. Phonology
The phonological side of the equation comprises representation and computation,
that is, a phonological structure organizing a set of symbols, and principles of
their manipulation. For reasons of space and relevance, the discussion of
representation is restricted to melodic primes (elements), while prosodic structure
is left out. A concrete illustration of the proposal will be based on the laryngeal
system(s) of Polish.
3
The privativity of phonological categories which is assumed here has been
argued for elsewhere and does not require additional argumentation (see, e.g.,
Avery 1996; Harris 1994, 2009; Honeybone 2002, 2005; Iverson and Salmons
1995; Lombardi 1991, 1995). Likewise, not much needs to be said about the
computation, especially within GP. In this theoretical model, segments are
composed of one or more elements and require licensing. Under insufficient
licensing conditions, for example, due to a particular prosodic context, segments
may be decomplexified (decomposition), while processes of spreading of
categories may lead to addition of elements to existing representations of
segments (composition). Below, I provide a simplified and rather uncontroversial
illustration of the four instances of processing operations: licensing,
decomposition, spreading, and composition, ignoring details which are irrelevant
to the discussion. (2a) shows voice assimilation as composition due to element
spreading. I follow, e.g., Gussmann (2007) in assuming that it is the laryngeal
element {L} that is responsible for the voice contrast among Polish obstruents.
The voiceless series is unmarked. In (2b), we observe a phenomenon of final
obstruent devoicing (FOD) in Polish as decomposition due to weak licensing in
the word-final context.
3
One should probably use the term ‘laryngeal sub-system’ here, equating ‘system’ with language
and allowing for a number of such sub-systems to be part of a larger system involving various
dimensions, for example, vocalic, place, manner, laryngeal, etc.
5
(2)
a. voice assimilation
b. FOD
weak licensing
p
r
O
Ç
+
b
a
m
a
Û
#
|
|
¯
< <<< L
L
prośba [pr
OÛba] ‘request’
maź [ma
Ç] ‘sticky substance’
Given that the voicing contrast in Polish is indeed expressed by the presence of a
privative element {L} in the representation of voiced obstruents, the processes
illustrated above can be described in the following way. If /
Ç/ is composed of
elements {x,y,z}, its voiced congener /
Û/ is one element more complex, that is,
{x,y,z,L}. In [pr
OÛba], the laryngeal element is spread from the following
obstruent. On the other hand, in [ma
Ç], {L} was present lexically, but delinked.
FOD is a case of decomplexification under weak licensing and turns {x,y,x,L}
into {x,y,z}. This is more or less the essence of privative analyses of such
phenomena. It should be emphasized that the unmarked (voiceless) obstruents do
not receive any further specification – they are interpreted as voiceless
unaspirated if {L} is not present in the representation.
What is more important for our discussion is how the categories receive their
phonetic definition in a substance-free phonology. The answer to this question
will not change much of the above analysis because we can always assume that
the set of symbols we use to discuss phonological phenomena, willy-nilly, must
already contain information as to what a given phonological category corresponds
to in the real world (of phonetics). Nevertheless, a possible way of looking for an
answer will be offered below.
The question of substance acquisition relates to one of the three main points of
interaction between phonology and phonetics (Kingston 2007). In discussions of
the definition of distinctive features, the typical problem is whether they are
articulatory, acoustic, or auditory, or in fact, whether they could holistically
involve all types. An imminent verdict on this issue is unlikely, and, as I will
argue below, unnecessary. From our perspective of the relationship between
phonology and phonetics a more important question seems to be whether phonetic
theory can model the emergence of the substance of the distinctive features.
Whether melodic primes are emergent and need not be postulated to be innate is
not a problem for substance-free phonology. The question that remains then is:
what is a feature, a categorical distinction, without substance? Our tacit
assumption at this stage will be that it is simply a decision to use an additional
contrastive dimension by assigning a new privative category to one of the
6
resultant contrastive series. The property will be given flesh by a systemic
interface with phonetics. For example, in the case of the /
Ç–Û/ contrast, where
{x,y,z} constitute the common denominator, it is a matter of introducing a fourth
element, or a fourth dimension of contrasts.
4
One of the functions of phonology is to define categorical contrast. In privative
models this boils down to a presence or absence of a particular property to
distinguish two segments. If no contrasts are used in a particular dimension, e.g.
laryngeal, then one series of obstruents is typically found – the voiceless
unaspirated, e.g. Hawaiian.
5
We may assume that such languages do not use any
laryngeal elements, a fact that will be represented below with a superscripted zero
next to C, which stands for an obstruent, that is, (C
o
).
In languages like Polish, or Icelandic, which have a two-way laryngeal contrast
among obstruents, that is, between a voiceless unaspirated and fully voiced for
Polish, and between a voiceless aspirated and voiceless unaspirated for Icelandic,
only one laryngeal element is used. In Polish, the marked representation involves
the presence of {L} in the voiced series (Gussmann 2007), while the neutral series
is ‘toneless’ (C
L
vs. C
o
). In Icelandic, on the other hand, the distinction is that of
(C
H
) for the aspirated series, and (C
o
), again, for the voiceless unaspirated one.
This, in essence, is the Laryngeal Realism view (Honeybone 2002, 2005; Harris
1994, 2009; Gussmann 2007). For completeness, one may add two other types of
systems: one with three and one with four contrastive series, which can also be
represented only with the two laryngeal elements mentioned above. Thai contrasts
three series /b,p,p
h
/, while Hindi has a four-way contrast /b,p,p
h
,b
H
/.
The privative representation of laryngeal contrasts in GP, which uses only two
elements for this dimension, including the Hindi case, appears to be compatible
with the finding of Lisker and Abramson (1964) that there are three major
phonetic categories which are utilized by languages, and which, as in the phonetic
descriptions above, can be quite elegantly illustrated by means of points or
regions along the VOT continuum. Firstly, there is full voicing, which can be
referred to as long lead or negative VOT. This property corresponds directly to the
phonological element {L} in Laryngeal Realism. Secondly, there are consonants
characterized by a short lag (voiceless unaspirated stops). This phonetic
realization seems to correspond most readily to the neutral obstruent (C
o
) which is
4
It need not be stressed that {x,y,z} are not real elements or dimensions. What exactly makes up
the fricatives is not relevant here.
5
In considerably fewer cases, it is the voiced series, e.g., in Yidiny (Keating, Linker, and Huffman
1983).
7
present in all four systems mentioned above. Finally, consonants may have a long
lag (voiceless aspirated stops). The element responsible for this distinction is {H}.
The relationships between the phonetic contrasts most commonly used in
languages and the GP elements is illustrated by the following graph.
6
(3)
closure
release
b
b
9~p
7
p
h
VOT:
lead
lag
t
C
L
C
o
C
H
Hawaiian
–
C
o
–
Polish
C
L
C
o
–
Icelandic
–
C
o
C
H
Thai
C
L
C
o
C
H
Hindi
C
L
C
o
C
H
/b
H
/ = C
L+H
Each system has to have the unmarked series /C
o
/. It should be noted, however,
that this involves a range of realizations from slightly voiced to voiceless
unaspirated. A laryngeal element, either {L} or {H} appears only if a two-way
contrast needs to be represented. Thai and Hindi utilize both elements, but the
latter language allows them both to be present in a single segment. [b
H
] is a
plosive which begins with a long lead and ends with a long lag. The English
system, for comparison, is phonologically similar to Icelandic in that it is assumed
to use {H}. However, its neutral obstruents are often realized with some voicing,
also called passive voicing. It is interesting to note, anticipating a little the
discussion of phonetic interpretation, that the passive voicing is possible only in
‘aspiration’ languages using {H} and impossible in L-systems. There seems to be
an asymmetry between voicing and aspiration languages (Lisker and Abramson
1964), in that fully voiced obstruents do not contrast with partially voiced ones,
while voiceless unaspirated can contrast with voiceless aspirated (e.g. Icelandic).
Thus, one contrastive region can be established on the VOT lead side, and two on
the lag side. This asymmetry may follow from the general phonetic fact that both
perceptually and articulatorily it is difficult to contrast fully voiced with slightly
voiced objects, and to control degrees of voicing.
6
The common practice is to use plosives in illustrations of VOT.
7
In aspiration languages, spontaneous voicing (also called passive voicing) may occur. The term
‘passive voicing’ will be explained further below.
8
So far, we see an almost biunique relation between the three phonetically
defined contrastive values along the VOT continuum corresponding directly to
three possible representations of stops in element theory, where full voicing in the
signal corresponds to the presence of {L}, aspiration relates to {H}, and voiceless
unaspirated objects are typically neutral. Thus, given the phonological marking
that is used in a given system it is directly obvious what phonetic values will be
used to express it, and vice versa: long VOT lead in the signal suggests the
presence of {L} in the representation of a given obstruent, while aspiration leads
us into thinking that it is connected with {H}. If the Laryngeal Realism view were
correct, phonological representation would always be unambiguously read off
from the spectrogram, and phonetic interpretation would be rather trivial.
However, the main problem with this model is that it does not work, at least for
one of the two major dialects of Polish, as will be shown below. First, let us look
at the phonetic side of the sound system.
2.2. Phonetics and phonetic interpretation
The phonetic side of the sound system presented in (1) contains principles of a
varying degree of generality rather than importance. Their role is strictly related to
phonetic interpretation of phonological representation. First, a distinction needs to
be drawn between universal phonetic principles and universal principles of
phonetic interpretation. These terms are not synonymous. The former relates to
physiology of speech and to phenomena which can be studied independently of
phonology. One example of such a principle is the general aerodynamics leading
to spontaneous vibration of vocal folds and its inhibition. The second term –
universal principles of phonetic interpretation – is ambiguous and misleading. It
suggests that phonological representation
8
contains universal instructions as to
how it should be pronounced. This would be compatible with the Laryngeal
Realism view presented above. However, it seems that the direction of motivation
may be the reverse. Phonetics provides options of phonetic interpretation, which
are selected or associated with particular phonological categories in a chiefly
arbitrary fashion. Secondly, phonetic interpretation is always system dependent,
that is, language specific, rather than universal. Nevertheless, if we understand
phonetic interpretation as a relation established between phonologically defined
8
It is possible that we also need a distinction between phonology proper, as a substance-free
computational module, and phonological representation (including lexical representation), in
which substance in the sense of established connections between subsegmental representation and
phonetic interpretation are present.
9
categorical contrasts and the phonetic contrastive regions, as in the case of the
three regions along the VOT continuum, we could also identify what appears to be
a universal principle of phonetic interpretation: the principle of sufficient
discriminability in production and perception. It is universal in the sense that most
known languages seem to follow it.
9
The universality in terms of production is
guaranteed among humans due to physiology. On the other hand, the same cannot
be said about perception. It may be possible to phonetically define a universally
potential maximal number of phonetic contrasts in a particular dimension. Just as
in the case of the VOT continuum, it is possible to define such contrasts in the
vowel space as well. The actual perception of speakers is always curtailed by the
particular system they have acquired. Thus, perception is to a great extent
language specific, unless we want to talk about potential and not the actual sound
systems.
Returning to the VOT contrasts, phonetics provides regions which allow for
minimal phonetic distance and therefore for discrimination. However, it is the role
of phonetic interpretation conventions to express the categorical distinctions
provided by phonology. This is where the universal principle of sufficient
discriminability, or better, sufficient phonetic distance comes into play, which is
to a great extent dependent on the number of contrasts demanding expression in a
given phonetic space (see, e.g., Liljencrants and Lindblom 1972). Thus, for
example, languages with a two-way laryngeal contrast tend not to select the
maximally dispersed phonetic categories: long VOT lead (fully voiced) contrasts
with short lag (voiceless unaspirated), rather than with long lag (voiceless
aspirated).
Sufficient phonetic distance does not only mean that contrasts need not be
maximized, it also means that there is something like a minimal distance. One
example of this has already been mentioned with respect to the VOT continuum.
Namely, no contrasts between full voicing and partial voicing (long and short
VOT lead) can be found (*/b–b
9/). Another interesting example concerns the
interpretation of the neutral obstruents (C
o
) in English and Icelandic in relation to
the marked congener (C
H
). Recall, that the neutral obstruents may be passively
voiced in English and tend not to be so in Icelandic. This fact coincides with the
phonological and phonetic robustness of aspiration in the two languages. Icelandic
aspiration is stronger than in English, both acoustically and in terms of
phonological behaviour. It tends to survive in more contexts than in English, and
9
The principle has been applied, for example, to the understanding of vowel systems (see, e.g.,
Schwartz, Boë, Vallée, and Abry 1997).
10
may be subject to temporal shifts rather than loss, a phenomenon called pre-
aspiration (Gussmann 1999). The observation which is relevant to this discussion
is that robust aspiration minimizes the chances for passive voicing, and vice versa.
It appears then, that the relation between the two obstruent series in these
languages observes something like a sufficient distance, where both the marked
and the unmarked series are subject to a coordinated variation. The following
graph attempts to express the main points of our discussion so far. Below, the
black circle denotes the marked obstruent series, while the white circle
corresponds to the unmarked congener. The dotted line between the black and
white circles indicates the sufficient phonetic distance, which is rather symbolic,
and it cannot really be measured in, e.g., temporal units as the graph may suggest.
The slight shift of the English marked-unmarked pair to the left in comparison to
Icelandic indicates that the aspiration is less robust and that the unmarked series
may be subject to passive voicing. It is passive, it will be recalled, because there is
no phonological category standing behind it. It is merely a systemic
interpretational phenomenon.
(4)
Phonetic distance and variation
closure
release
Polish
a.
/b/ vs. /p/
Icelandic b.
/p
h
/ vs. /p/
Germanic
English
c.
/p
h
/ vs. /b
9/
t
phonological symbols: C
L
C
o
C
H
VOT:
lead
lag
Universal and language specific principles of phonetic interpretation do not seem
to have a clear boundary, especially with respect to certain types of segments, for
example, obstruents. Let us dwell a little on the question of the aerodynamic
conditions inducing vocal fold vibration (voicing) in order to be able to show how
this universal phonetic principle is affected by systemic (language specific)
considerations.
The vibration of vocal folds occurs under special aerodynamic conditions
involving a number of articulatory parameters. The desired effect of these
articulatory settings is to achieve a sufficient drop in air pressure and air flow
between trachea and pharynx (Chomsky and Halle 1968; Halle and Stevens 1971).
The drop in air pressure is inhibited in segments which are produced with some
narrowing in the vocal tract because occlusion leads to intra-oral air pressure
11
build-up. These simple physical facts are responsible for the so called universal
markedness tendency for vowels and sonorant consonants to be voiced and for
obstruents to be voiceless. In phonological descriptions, these simple aerodynamic
facts are often expressed by the use of the following default rules (e.g., Gussmann
1992: 43; Rubach 1996: 77,80).
(5)
a. [sonorant]
→
[+voice]
b. [obstruent]
→
[–voice]
While (5a) seems to be overwhelmingly correct – vowels and sonorant consonants
are typically voiced, obstruents seem to defy the supposedly phonetically natural
rule in (5b). First of all, sonorant voicing is generally considered spontaneous,
which has lead to proposals that it should not be expressed phonologically by
means of any feature or element. This is also the position of a number of privative
feature frameworks, including the Laryngeal Realism view and the model
presented here.
As for obstruents, under certain articulatory and contextual conditions they also
may be spontaneously voiced (Westbury and Keating 1986). We may generally
describe these conditions as lenis articulation
10
and voiced environment, that is,
adjacent vowels or sonorant consonants. Due to the fact that, unlike in sonorants,
such voicing is dependent on the environment, instead of spontaneous, the term
passive voicing is often used to refer to this situation (e.g., Kohler 1984).
Westbury and Keating (1986) note an interesting paradox about some of the
languages possessing only one series of obstruents. Recall, that this concerns the
segments which we symbolize as C
o
, that is, laryngeally unspecified ones, which
are typically realized as voiceless unaspirated. This voiceless articulation is
maintained also in contexts (voiced environment) in which spontaneous voicing
would be phonetically more natural. It would seem then, that the default rule (5b)
above may in some cases be phonetically unnatural. Westbury and Keating
acknowledge that this lack of voicing is due to ‘more powerful principles’, for
example, a systemic tendency to maintain the phonetic similarity among the
positional allophones. Clearly, these more powerful principles override natural
phonetics and should be viewed as stemming from the interaction between
10
These include, for example, relatively short closure, contracting the respiratory muscles,
decreasing the average area of the glottis and / or tension of the vocal folds, decreasing the level of
activity in muscles which underlie the walls of the supraglottal cavity, actively enlarging the
volume of that cavity, etc.
12
phonetics and phonology. These are phonologically dependent decisions on
phonetic interpretation of segments.
To conclude this part of our discussion. Thus far, it is clear that phonetic theory
alone turns out to be insufficient in the study of sound systems. Likewise, without
phonetics providing predictable contrasting regions, substance-free phonology
would be equally lacking. It appears that phonetic interpretation is not a
phonological instruction. Rather, it is an interface phenomenon. Phonology
provides the number of contrasts, while phonetics provides the possible phonetic
contrasts. The relation between the two could potentially be quite arbitrary,
11
though complying to some principles. Below, I relate another argument in favour
of the arbitrariness of the relation between phonology and phonetics in sound
systems.
3. Sandhi voice assimilation in Cracow-Poznań Polish
Polish divides into two dialect groups with respect to voicing: the Cracow-Poznań
(CP) and Warsaw (WP). The phonetic and phonological facts in these dialects
seem to be generally identical, except for the assimilation phenomena across word
boundary, the so called sandhi voicing. Thus, both dialects have a two-way
laryngeal contrast between obstruents involving fully voiced and voiceless
unaspirated congeners. Both dialects boast the same processes, such as final
obstruent devoicing (FOD) and word-internal voice assimilation (VA). However,
the voice assimilation phenomena in the external sandhi context are markedly
different. In CP, a word-final obstruent becomes voiced before any voiced
segment beginning the following word, that is, before a voiced obstruent, a vowel,
or a sonorant consonant. In WP, on the other hand, voice assimilation occurs only
if the following word begins with a voiced obstruent. The facts are independent of
the lexical representation of the final obstruent.
The data below show lexically voiced and voiceless stops in the context before
another word beginning with a vowel (_V
+v
), a sonorant consonant (_S
+v
), a
voiced obstruent (_C
+v
), and a voiceless obstruent (_C
–v
). The superscripted voice
values here a purely phonetic.
11
The arbitrariness of these relations deserves a longer discussion which will have to be reserved
for another occasion. Suffice it to say that, like in the acquisition of vocabulary, the relation
between the concept and the phonological form that expresses it is not at all arbitrary to the
learners of their language. It is rather arbitrary for the linguist.
13
(6)
WP
CP
a.
kwiat akacji ‘acacia flower’
t-a
d-a
__V
+v
b.
kwiat róży ‘rose flower’
t-r
d-r
__S
+v
c.
kwiat bzu ‘lilac flower’
d-b
d-b
__C
+v
d.
kwiat paproci ‘fern flower’
t-p
t-p
__C
–v
e.
sąd apelacyjny ‘court of appeal’
t-a
d-a
__V
+v
f.
sąd rodzinny ‘family court’
t-r
d-r
__S
+v
g.
sąd wojenny ‘court-martial’
d-v
d-v
__C
+v
h.
sąd karny ‘criminal court’
t-k
t-k
__C
–v
Let us begin with some general observations. Firstly, the fact that the lexical
origin of the word-final obstruent is irrelevant in (6) above – they behave
uniformly – suggests that we are dealing with some kind of neutralization in that
context. In terms of privative representations discussed earlier, the common
denominator of the two lexical representations and the target of sandhi voice
assimilation is an unmarked obstruent (C
o
).
12
Secondly, as the data above show,
both WP and CP have a sandhi assimilation. However, in WP it is limited to the
context in which the trigger of assimilation is a voiced obstruent. In other words,
somehow voicing in obstruents is distinguished from that in sonorants in WP, but
not in CP. The entire analysis of CP sandhi voicing is dependent on how this
distinction is expressed in phonological models.
In binary feature models in which sonorants, including vowels, also carry
[+voice], the distinction between sonorant voicing and that in obstruents cannot be
expressed in the representation. It is written in in the assimilation (VA) rule. In
WP, the VA rule specifies that [+voice] spreads only from obstruents, while in CP
this feature spreads irrespective of the type of segments in which it resides. It is
clear, that sonorants must have [+voice] in their representation, otherwise, the VA
rule in CP would not be able to refer to them as triggers. What is not so clear is
how phonology distinguishes between an autosegmental feature [+voice] that can
spread from one that does not, as is the case in WP.
In most privative models, however, sonorants do not have a feature [+voice],
and, for example, in Laryngeal Realism they do not receive this property in the
process of derivation, or through defaults either. Sonorants are non-specified for
12
In phonological models using a binary feature system, where Polish /p/ has [–voice] and /b/ has
[+voice], the delaryngealized common denominator is a third type of segment, one that never
surfaces. It is available for assimilation processes only at a particular stage of derivation. Rules of
assimilation which aim to produce a different voice value than the default one on that obstruent
must hurry.
14
voice because they are voiced spontaneously. If no phonological category stands
behind sonorant voicing, then, obviously, they cannot be triggers of phonological
voice assimilation. This has good and bad consequences. The positive outcome is
that we have a representational means of distinguishing between the voicing in
obstruents as due to the presence of a phonological category, and that in sonorants
in which it is spontaneous (phonetic). Thus, the VA rule in WP could simply say:
‘spread Lar’, where Lar stands for a laryngeal category, in our terms the element
{L}. The negative side of this analysis, which is compatible with Laryngeal
Realism, and follows the representation of voice in Polish proposed in Gussmann
(2007), is that it seems to work only for WP. In that dialect C
o
must be interpreted
as voiceless unaspirated unless it receives the element {L} by spreading. Since
only obstruents have this element, the analysis predicts that sandhi voicing will be
restricted to that single context, e.g., kwiat bzu [kf
j
ad bzu] ‘lilac flower’. On the
other hand, the scope of the CP sandhi voicing is inexpressible in an L-system
proposed by Gussmann. In CP, it will be recalled, both obstruents and sonorants
voice the word-final neutralized obstruent (C
o
), e.g., kwiat róży [kf
j
ad ru
ZÈ] ‘rose
flower’, kwiat akacji [kf
j
ad akatsji] ‘acacia flower’.
A solution to this problem was proposed in Cyran (2011, 2012) which, if
correct, has far reaching consequences for Laryngeal Realism and for the
relationship between phonetics and phonology. In short, it is proposed that the two
dialects of Polish have opposite lexical representations of voicing with respect to
which series of obstruents is marked with a laryngeal element and which one is
neutral. This entails markedly different phonetic interpretation rules, which lead to
practically identical phonetic facts (observable data), but with a notable exception.
Let us first look at a relevant graph illustrating the representational difference
between CP and WP.
(7)
Polish dialects in Laryngeal Relativism
closure
release
a. Warsaw
C
L
vs. C
o
b. Cracow-Poznań
C
o
vs. C
H
b
p
t
VOT:
lead
lag
C
o
in WP cannot be interpreted as voiced under any other conditions than the
presence of {L}. This is the effect of being the neutral series in an L-system. CP,
on the other hand, is an H-system, where the marked series is on the lag side of
15
the VOT continuum. C
o
in CP is only technically (representationally) the same
object as C
o
in WP. However, its systemic situation is different. The principle of
sufficient phonetic distance between the two series enforces its phonetic
interpretation to be fully voiced (long lead). Thus, we are dealing here with a
system in which C
o
must be passively voiced, and that voicing – for reasons to do
with phonetic distance – must be more robust than what we observe in, for
example, English. For this reason this case is given a separate name in Cyran
(2011), namely, enhanced passive voicing. It is not impossible that this realization
has another cause apart from systemic interpretation. The close contact between
the dialects – they belong to one language – enforces phonetic uniformity in the
interpretation of the two series as voiceless unaspirated versus fully voiced.
13
Arguments in favour of the disparate phonological representation of the voice
contrast in the two dialects of Polish are given in the references quoted above. For
our purposes it is important to see how this ‘mirrored’ system allows us to
understand CP sandhi voicing and what this analysis tells us about our main
problem of the relation between phonology and phonetics. Let us begin with the
former aspect and note that when an obstruent C
o
in a given system is interpreted
as voiced, or even fully voiced, then its voicing is phonologically speaking no
different from that in sonorants. Namely, there is no phonological category
standing behind this voicing that could be manipulated by phonological
computation. Predictably, such obstruents should behave on a par with sonorants,
including vowels. There is one difference though. The passive voicing in
obstruents, unlike spontaneous voicing in sonorants, requires phonetically voiced
context. For this reason, C
o
in an H-system may not be passively voiced in, e.g.,
word-final context, and is pronounced without voicing. This seems to be true in
CP, where absence of passive voicing word-finally, due to absence of voiced
context, can be easily confused with a real phonological process of FOD. C
o
in
Cracow-Poznań Polish is voiceless for a different reason: it is an absence of
passive voicing rather than devoicing. We predict then, that in the face of a voiced
segment which begins the following word, and under the condition of adjacency
(no pause intervening), the word-final C
o
in CP should be interpreted phonetically
as voiced, not only in front of a voiced obstruent, but also in front of sonorants.
This is what happens in the celebrated phenomenon of Cracow-Poznań sandhi
voicing.
13
Shifts in laryngeal systems due to language contact are not unknown (see e.g., Honeybone
2002). Inter-dialect contact is surely a more powerful phenomenon.
16
To summarize. The CP facts fall out only if an H-system is assumed, in which
C
o
can be voiced without an addition of a phonological category which would be
responsible for this property, e.g., element {L}. It will be recalled that C
o
in WP,
which is an L-system, cannot be voiced in any other way than by getting {L}.
Passive voicing is possible only in H-systems. C
o
in CP requires phonetically
voiced context in order to be voiced, hence, word-final non-voicing, and sandhi
voicing in front of phonetically voiced segments, that is, vowels, sonorant
consonants and other passively voiced obstruents. CP sandhi voicing is not due to
spreading of [+voice] or any category for that matter. It is an interpretational and
obligatory phenomenon in that system. No special rule is even needed. Below, in
the concluding section, I present some consequences of this analysis on the way
we should perceive the relation between phonetics and phonology.
4. Some consequences of Laryngeal Relativism
The approach presented above, in which the relations between phonetically
provided contrastive regions along the VOT continuum and the phonological
categories can be established with such a degree of variation as to allow for
systems with exactly opposite representations to yield identical phonetic effects in
word phonology will be called Laryngeal Relativism. It defies and rejects
Laryngeal Realism in which phonetic facts have a direct translation into the
phonological representation standing behind them, and vice versa. In the new
approach, it is assumed that phonetic facts are no guarantee of direct access to
phonological representation and the whole architecture of a given system.
To put it in more specific terms, C
H
has more variation in the way it may be
phonetically interpreted than it was believed before on the basis of, e.g., Icelandic
and English. {H} may also mark a voiceless unaspirated series. The cause of that
variation is that the laryngeal categories do not exist outside particular systems in
which the main principle of the distribution of phonetic contrasts is sufficient
phonetic distance, and the connections between phonological and phonetic
categories are characterized by a fair degree of arbitrariness.
The neutral series of obstruents (C
o
) has no universal (default) interpretation as
voiceless unaspirated. The discussion of the Polish facts allows us to say that the
scope of the interpretation of C
o
is from fully voiced to voiceless unaspirated.
Thus, in the two cases above, we break with the unwarranted biuniqueness
between representation and phonetic effect. These relations are system specific
and subject to shifts, as illustrated below.
17
(8)
CP
b ~ p
C
L
C
o
C
H
WP
b ~ p
The question of the scope of C
L
is a little more complicated. So far, we have no
evidence on the basis of which we could claim that the presence of {L} does not
guarantee an interpretation of obstruents with a long VOT lead (full voicing). It
seems that {L} must be interpreted as long lead, but not the other way round.
Long lead in the phonetic facts does not yet guarantee that the element {L} is
responsible for it, because we saw that, for example, the CP facts are better
understood if we assume that the long lead is an interpretation of C
o
in an H-
system.
The above variation leads us to yet another interesting observation about the
way phonetic interpretation works. It transpires from this discussion that phonetic
interpretation seems to take into account the entire segment rather than individual
features. This is visible not only in the interpretational shifts illustrated in (8), but
also in the way sonorants are interpreted as opposed to obstruents with respect to,
for example, voicing. The former are universally voiced (spontaneous), while the
voicing in the latter (passive) is system dependent, possible only in H-systems,
and context dependent – a voiced environment is required for passive voicing.
It is quite clear that in Laryngeal Relativism, neither phonological
representation can function without phonetic interpretation rules, nor surface
phonetic facts can give us an unmistaken clue as to the phonological
representation and the architecture of the system. So the two domains cannot
function independently in a meaningful way, yet they are independent and the
relations between them are to a large degree arbitrary. Of course, there are natural
restrictions on this arbitrariness, for example, the three phonetic regions of
stability used for laryngeal contrasts do not leave much leeway for choice.
There is a lot more that can be said about the arbitrariness of the relationship
between phonological and phonetic categories. Firstly, it may be more readily
accepted if we realize that such relations pervade language. For example, except
for cases of sound symbolism one would not normally look for a direct link
between the phonological string /k
h
œt/ and the concept it signifies. Such relations
are arbitrary, and yet in a sense obligatory when looked at from the perspective of
language acquisition. Learners are not faced with complete arbitrariness and they
do not construct these relations anew. They have no choice if they happen to be
18
learners of English. Language acquisition is not a case of language birth. It is
more of a language reconstruction.
A similar situation may be assumed to take place with respect to acquisition of
phonological systems. What is given as input at language acquisition stage is
phonetic facts, that is, one side of the coin. The system, which will take these facts
into account and link them to phonological representation through a set of
phonetic interpretation rules, must be worked out.
More importantly for our discussion though, the arbitrary relation between
phonetics and phonology, if true, forces us to say that substance may be emergent,
or derivative, and not part of UG. Note that earlier, this was a mere speculation, or
a hypothesis on our part. Now, in the face of the analysis of CP sandhi voicing
and the mirrored representations in the two Polish dialects, we must accept it as a
necessary view. Note that if the elements {L} and {H}, defined as long lead and
long lag were innate and part of UG, as Laryngeal Realism and standard Element
Theory in GP would have it, then shifts of the sort illustrated in (8) would be
impossible. Under this view, a shift in phonetic categories would also have to
entail a shift in phonological representation (Honeybone 2002). The above shifts
are made possible within Laryngeal Relativism, in which the connections between
phonological and phonetic categories are acquired. They are also subject to
change through language contact or due to other historical developments. A shift
in phonetic categories need not entail a shift in phonological representation. There
is also the possibility that only the phonetic interpretation rules have changed.
As for the debate on the nature of distinctive features, once we accept the
arbitrary nature of the relationship between phonetic and phonological categories,
the issue becomes spurious. To emphasize this point further it should be admitted
that the VOT continuum used in this discussion is not the only possible way of
defining phonetic categories to do with laryngeal contrasts. One could think of
equally successful articulatory definitions operating with such properties of vocal
folds as ‘stiff’, ‘slack’ and ‘spread’ (e.g., Halle and Stevens 1971), or articulatory
dimensions ‘glottal tension’ and ‘glottal width’ (Avery and Idsardi 2001). The
articulatory, acoustic or indeed auditory nature of the phonetic substance does not
change the concept of the sound system architecture described in this paper. Even
under the extreme assumption that phonology is substance-free and phonetics is
almighty in the study of sound patterns, the two domains must be studied together
through the medium of phonetic interpretation rules. I believe this is what Stieber
meant.
19
References
Avery, Peter
1996 “The representation of voicing contrasts”. PhD Dissertation. University of Toronto.
Avery, Peter – William Idsardi
2001 “Laryngeal dimensions, completion and enhancement”, in: Tracy Hall (ed.), Distinctive
feature theory, 41-70. Berlin and New York: Mouton de Gruyter.
Brockhaus, Wiebke
1995 Final devoicing in the phonology of German. Tübingen: Niemeyer.
Chomsky, Noam – Morris Halle
1968 The sound pattern of English. New York: Harper & Row.
Cyran, Eugeniusz
2011 “Laryngeal realism and laryngeal relativism: Two voicing systems in Polish?”, Studies
in Polish Linguistics 6: 45-80.
2012 “Cracow voicing is neither phonological nor phonetic. It is both phonological and
phonetic“, in: Eugeniusz Cyran – Henryk Kardela – Bogdan Szymanek (eds.), Sound,
structure and sense. Studies in memory of Edmund Gussmann. Lublin: Wydawnictwo
KUL.
Gussmann, Edmund
1992 “Resyllabification and delinking: the case of Polish voicing”, Linguistic Inquiry 23: 29-
56.
1999 “Preaspiratin in Icelandic: unity in diversity”, Studia Anglica Wratislaviensia 35: 161-
179.
2007 The phonology of Polish. Oxford: Oxford University Press.
2009 “Alexander J. Ellis on modern Icelandic pronunciation”, Studia Linguistica
Universitatis Iagellonicae Cracoviensis 126: 47-59.
Hale, Mark – Charles Reiss
2000 “Substance abuse and dysfunctionalism: current trends in phonology”, Linguistic
Inquiry 31: 157-169.
2008 The phonological enterprise. Oxford: Oxford University Press.
Halle, Morris – Kenneth N. Stevens
1971 “A note on laryngeal features”, MIT Quarterly Progress Report 101: 198-212.
Harris, John
1994 English sound structure. Oxford: Blackwell.
2009 “Why final obstruent devoicing is weakening”, in: Kuniya Nasukawa – Philip Backley
(eds.), Strength relations in phonology, 9-45. Berlin and New York: Mouton de Gruyter.
Honeybone, Patrick
2002 “Germanic obstruent lenition: some mutual implications of theoretical and historical
phonology”. PhD Dissertation. University of Newcastle upon Tyne.
2005 “Diachronic evidence in segmental phonology: the case of laryngeal specifications”, in:
Marc van Oostendorp – Jeroen van de Weijer (eds.), The internal organization of
phonological segments, 319-354. Berlin and New York: Mouton de Gruyter.
Iverson, Gregory – Joseph C. Salmons
1995 “Aspiration and laryngeal representation in Germanic”, Phonology 12: 369-396.
20
Keating, Patricia – Wendy Linker – Marie Huffman
1983 “Patterns in allophone distribution for voiced and voiceless stops”, Journal of Phonetics
11: 277-290.
Kingston, John
2007 “The phonetics-phonology interface”, in: Paul de Lacy (ed.), The Cambridge handbook
of phonology, 401-434. Cambridge: Cambridge University Press.
Kohler, Klaus
1984 “Phonetic explanation in phonology: the feature fortis/lenis”, Phonetica 41: 150-174.
Liljencrants, Johan – Bjorn Lindblom
1972 “Numerical simulation of vowel quality systems: the role of perceptual contrast”,
Language 48(4): 839-862.
Lindblom, Bjorn
1986 “Phonetic universals in vowel systems”, in: John Ohala – Jeri Jaeger (eds.),
Experimental phonology, 13-44. Orlando: Academic Press.
Lisker, Leigh – Arthur Abramson
1964 “A cross-language study of voicing in initial stops: acoustical measurements”, Word 20:
384-422.
Lombardi, Linda
1991 “Laryngeal features and laryngeal neutralization”. PhD Dissertation. University of
Massachusetts, Amherst. Published (1994), New York: Garland.
1995 “Laryngeal features and privativity”, The Linguistic Review 12: 35-59.
Ohala, John
1990 “There is no interface between phonetics and phonology”, Journal of Phonetics 18:
153-171.
Ploch, Stefan
1999 “Nasals on my mind. The phonetic and the cognitive approach to the phonology of
nasality”. PhD Dissertation. SOAS, London.
Rubach, Jerzy
1996 “Nonsyllabic analysis of voice assimilation in Polish”, Linguistic Inquiry 27: 69–110.
Schwartz, Jean-Luc – Louis-Jean Boë – Nathalie Vallée – Christian Abry
1997 “The dispersion-focalization theory of vowel systems”, Journal of Phonetics 25: 255-
286.
Westbury, John R. – Patricia Keating
1986 “On the naturalness of stop consonant voicing”, Journal of Linguistics 22: 145-166.
Stieber, Zdzisław
1955 “Na marginesie dyskusji fonologicznej [On the margin of phonological discussion]”,
Rozprawy Komisji Językowej Łódzkiego Towarzystwa Naukowego 2. 73-74.